Just out of curiosity, I decided to benchmark the time difference between loading a web graphic directly from a web server vs. loading it from my ISPs cache server. Please don't draw any generalization from this one isolated test, your result could vary substantially depending on what sites you access and the type and configuration of your ISP's caching devices.

Test scenario:

Provider: Bellsouth.net ADSL
Cache server: fcs00asm.asm.bellsouth.net
Test page: A 20KB graphic from cnn.com's home page

Not-Cached:

Cached:

Observations:

• There was little difference in the average response time
• Non-cached connection was actually faster for some trials
• Non-cached response times were more variable
• Cached response times were very consistent
• Non-cached connection reused the same TCP connection for each page load
• Cached created a new TCP connection each time the object was requested, and tore it down as soon as the object was loaded

Other factors to consider:

Bellsouth currently has transparent cache redirection disabled, so the cache server likely had almost zero load. Given this fact I had expected the cached connection to be significantly faster than non-cached. In the past (when the cache servers were in full production) I observed significantly higher TCP connect times (sometimes > .500s). Given that the cache servers appear to create a connection for each object retrieved, I suspect this may be the reason that I've seen poor cache performance in the past--but this is just a guess for now.

Raw packet traces are available here:

CnnNoCache
CnnCache